The document has a lot of new content that will be published separately, including some of my quick highlights and opinions of the field in the background section.pic.twitter.com/G69AjFpRed
Эта настройка позволяет добавлять в твиты информацию о местоположении, например название города и точные координаты, на веб-сайте и в сторонних приложениях. Вы можете удалить сведения о местоположении из своих твитов в любое время. Подробнее
The document has a lot of new content that will be published separately, including some of my quick highlights and opinions of the field in the background section.pic.twitter.com/G69AjFpRed
Here are the optimization viewpoints of the ReLU, sigmoid, and softmax layerspic.twitter.com/VWI1SXvIVf
And we've been working on a new way of quickly prototyping optimization layers with cvxpy and @PyTorch that enables most of my PhD work to be re-implemented in a few lines of code. This is just a preview and we'll be finishing up and releasing the software package soon
e.g. to start easy, here's how you would implement the optimization viewpoint of the ReLU, sigmoid, and softmaxpic.twitter.com/tVHdn6lW6R
Here's a complete example of how the OptNet layer can be implemented now:pic.twitter.com/yOYqfrOZR1
And here's how you can learn a polytope or ellipsoid in the latent space of your model, and you can easily add any other parameterized convex constraint you wantpic.twitter.com/bJD2kVGGmG
Congrats man 

May I ask what is your writing workflow for this. Like the editor used, programs, etc
For writing I'm becoming a dinosaur and mostly just live in emacs/auctex/git and use powerpoint/matplotlib for diagrams
That’s great, I too use AucTex with Reftex and pdf-tools. Seems like a match made in heaven.
oo I've been wanting to start using reftex/pdf-tools too
It looks very interesting. Let us know at what time and room will you defend, some of us may want to be there!
Thanks! It's on Thurs May 2 at 11am in NSH 4305https://www.cs.cmu.edu/calendar/thu-2019-05-02-1100/computer-science-thesis-oral …
nice try, but you'll never know who is attacking.
Here's to hoping your defense is adversarially robust!
Fantastic work! Good luck :)
looks like great work toward allowing recent NN approaches to be more directly applied to more things by providing a uniform process for incorporating domain knowledge, especially where limited training samples are available, or when computational limits make models desirable.
Вероятно, серверы Твиттера перегружены или в их работе произошел кратковременный сбой. Повторите попытку или посетите страницу Статус Твиттера, чтобы узнать более подробную информацию.